Well, I decided to give my first attempt at processing a vinyl LP that I recorded to WAV using Audition 2.0. It's an older album that got lots of play time by Meat Loaf called Bat Out Of Hell. I could hear a few clicks, but overall I thought the album sounded pretty darn good.

So at first I do a little looking for visible clicks using the spectral frequency display. I can see some so I give the "Auto Click/Pop Eliminator" a try. The three deault presets "light, medium and heavy reduction" all seem to eliminate the visual spikes pretty well. There is no indication of how many clicks were prcessed. The album still sounds good.

So I undo these processes and decide to start reading about the full featured "Click/Pop Eliminator". This seems to be a very different beast, not even including the "noise threshold" or "complexity" sliders that were present in the "Auto" version. So I read the entire printed and online manual about this section. There is some helpful advice, but many things are still glossed over. So I'm really not sure if I used the tool correctly. But I left "sensitivity" and "discrimination" unchanged at 14 and 20. I chose a preset of "Medium Amplitude Audio". This set new numbers for the "Threshold, Detect and Reject" levels. It also chose some check boxes and set "Pop Oversamples" and "Run Size" to 12 and 20. I guess I could have just run the eliminator at this point, but I couldn't resist clicking on the "Auto Find All Levels" button. This took several minutes, but when it was finished, all of the "Threshold, Detect and Reject" levels had been changed. I don't fully understand this, but I assume that the "Auto Find All Levels" process attempts to fine tune the settings specifically for my recording, instead of just using the "Medium Amplitude Audio" presets. However, the check boxes and "Pop Oversamples" and "Run Size" remain unchanged at 12 and 20.

So now I'm thinking, OK, Audition has set the best parameters that it can, let's give this a try! Remember, I'm applying this to the entire WAV file of about nine songs. Several minutes later I'm informed that 49,962 clicks were corrected and 67,571 rejected! I can't believe this. I heard and saw a few clicks in the spectral frequency display, but not 49 thousand. So I assume that the auto find settings must have been way too sensitive and that my recording must now sound like swiss cheese. But nooooooooooo, it sounds great! I'm still completely unnerved by the fact that so much processing was done to my file. But I also kind of like the idea that maybe Audition can do a better automatic click/pop removal than I could ever do.

Here are some questions:

1. Do you think 49 thousand click removals was excessive? I can provide the exact threshold, detect and reject settings that were used if this would be helpful. Another curiousity I noticed was that if I just clicked on the button "Find Threshold Levels Only", it returned different Threshold levels than what the "Auto Find All Levels" button returned.
2. Do people prefer to use the simpler "Auto Click/Pop Eliminator" rather than the "Click/Pop Eliminator"? I don't really understand the chosen names for these two options. They both eliminate clicks automatically. The second just has more variables that you can control.
3. Did I use the "Click/Pop Eliminator" correctly? Do people like to use that "Auto Find All Levels" button? I can only assume that button should tailor the click removal process more accurately to your specific recording than using one of the four presets.

I tried one other evaluation after the big cheese carving incident. I took the original WAV file, inverted it and mix pasted it into the swiss cheese copy. This was supposed to allow me to hear the pops that were removed. Well it worked pretty neat. In spectral frequency view I could only see thousands of narrow spikes. Upon listening, there was no music, just frequent, very soft clicks. Many were almost inaudible, and if present in the original music, must be completely masked by the normal music volume. Here's one final question:

4. Do you think that all these excised clicks are probably real? Or would the spectral frequency display of my mix paste file incorrectly show everything as narrow, click like spikes even if nomal music frequency had been accidentally removed by Audition?

Oh my gosh!

The short answer to your last question is that the process removed ‘real' material but much of it is not clicks or other impulse noise. Fortunately much of that doesn't really matter, in a sense similar to the fact that the difference between the original and a good mp3 encoding doesn't matter -- but sometimes it does. In fact, it is quite easy to totally destroy a recording with the wrong parameter values.

For a limited discussion of some considerations, you might look at this thread. I wrote some instructions for investigating the differences near the end of the thread.
http://www.audiomastersforum.net/amforum/i...pic,5441.0.html

I think this Auto Click/Pop Eliminator must be something new, not in the version of the program I have. Since the Click/Pop Eliminator is endlessly complex, this newer addition is most likely an attempt to make things easier for those who don't want to get too involved. If you do, there were many long threads discussing options and techniques on the Syntrillium forum (original program developers), available through the archive button at the top os the above forum.

Good questions. Sorry I don't know enough to be of any real help as I'm still just beginning to learn about Audition, and I wasn't a Cool Edit user before Adobe took over.

I did just want to point out that by my calculations, if you recorded "Bat Out of Hell" as 16-bit 44.2 kHz data, you have about 246 million samples of data. So 49,000 "clicks" removed is not a huge number relatively speaking, especially if you started with a somewhat noisy LP to begin with. OTOH, I would think Audition removed a lot of "clicks" that either weren't really clicks, or were impulse-type noise that would not be noticeable if left in.

So the conservative approach would be to back off on the settings until you no longer have any audible clicks. A more efficient way to work would be to pick out the noisiest (click-wise) track, extract it and save it as a new .wav file, then work with it to find the best settings, and apply them to the whole album.

Sorry, not a direct answer to your question, because I still use Audition 1.5 and "Clickfix" plugin (http://www.jdklein.com/clickfix) , so I have no idea how 2.0 handle click/pop.

But I can provide some information as your reference, after making 20GB of lossless FLAC, I got some "average" numbers.

A normal vinyl LP, not too noisy, after cleaning/washing, the clicks should below 2000 (5 mins), of course how many clicks depends heavily on the status of the LP, if it's in very good condition, I got less than 1000 clicks, if the number is too large, it may cause by the high recording level, not really clicks.



No one can answer this question because it depends.

Try all the buttons, and compare all the processed sound clip with your ears, and you'll got the answer.

You should only care about how much "real" music being removed, if you really want to preserve the most (like me), it's quite easy, to lower the filtering parameters threshold, and compare with the non-processed original.

Buy a better sound card and speaker will help a lot. I didn't recommend Sound Blaster Live or Audigy, they sound awful.

I can't add much that hasn't already been said. I don't use Audition myself - I have the older CoolEdit 2000 with the audio cleanup plugin. But my understanding is that Click/Pop Eliminator algorithms are the same. (They certainly were back in the days when Audition was called CoolEdit Pro).

In my experience "Auto Find All Levels" usually arrives at settings that are too sensitive and frequently result in a declicking operation that audibly degrades the music. It would seem that this did not happen in your case. There are two possible explanations that occur to me:
1. You were lucky with that particular album.
2. Maybe the algorithms have been improved in Audition 2.0.
Either way, I don't think you should rely on the efficacy of Auto Find All Levels before you've done a lot more experimentation.

As for whether 49,000 clicks over a whole album is a lot: I'd say that's fairly typical of what the Click/Pop Eliminator finds on an album with typical amounts of wear. I personally think that the majority of them are probably phantoms and need not be removed, but as Andy says, provided there is no audible degradation, it doesn't matter if phantoms do get removed.

One final word of advice: automatic declickers rearely perform well with big clicks. They often replace them with "thuds" and "splats". So it's a good idea to do a manual pre-pass to remove the big clicks by hand before you let an auto declicker loose.

There is one difference in Audition 2.0 .

When in spectral view you can select the frequency range on which the declick algorithm is applied.

By manual do you mean select a small segment of waveform that contains the big click and then do an "automatic" operation like fill single click, or do you actually mean to excise the click manually? If completely manual, is there a way to manually replace samples that would approximate the waveform?

Fill single click now is generally though of as manual. You must select exactly what you want, you must perform the operation (via program controls, as you do everything on a computer). In five years of writing about this topic I haven’t come up with a good word for the mode of having the computer do as much as possible, that is to say, a word that doesn’t have confusing implications for some readers.

Automated declicking is declicking in batch mode, but some people have difficulty with the fact that it is the clicks being batched, not a number of different files done via a single command. Then again, automated isn’t really completely automated in most programs. You must select the audio (entire file or some limited section thereof) and you must assign the values to some number of operational parameters. I suspect the only solution is to aleays be detailed enough to be sure that everyone interested get on the same page.

In Audition, there are detection parms and correction parms, independent of each other. The same correction parms work differently in auto and manual modes. Also, for manual declicking, selecting a very small section, often only one sample, often gives the best results, but sometimes it is necessary to select considerable more than the click itself. Most often more of that selection needs to be after the click than before.

Often the correction needs to be repeated some number of times, often, but not always, without changing the selection. Sometimes more than one set of correction parameters are needed on a single click to get the best results.

What/how much is removed or modified for the count or one click depends very much on the correction parameters and the audio material, so the number of clicks that Audition gives you is not meaningful in any absolute way. The results on the music of an given fix also depends quite a bit on the immediate nature of the music, the context in which the click fix is applied. I've had combined counts that may have exceeded two million removals/fixes for multiple declicking runs on one album. The results were definitely better than the beginning.

In the first two years of LP transfers, I probably did a couple hundred hours of experimentation, often repeating declicking runs on sample tracks, using different parameter values, to find what worked best. The ultimate goal had to be to become familiar enough with results in many different kinds of music. The same parameter values could not be used just anywhere; they had to be matched to the material.

I don't remember that much about having the program find the levels to be used for detection. I do know that I developed some presets for particular circumstances that used particular values for threshold, detection, and rejection -- the program was not allowed to change those values before the run. Other people reported on their own presets, That does not mean that the program did not sometimes do a better job when allowed to run free.

Part of the difficulty is that the interactions are so complex. What works very well on some material might really destroy other material, sometimes in the same recording. Although I got a fair number of good transfers out of this, all the effort led me to eventually adopt other tools which I can use without so much hassle.

ClickFix is cheap and is sometimes quite useful. It isn't a panacea in very many circumstances, although some users seem to thing so. The Sonic Foundry Noise Reduction 2 plug in does most of my declicking these days, but I use ClickFix, and sometimes CoolEdit declicking, on parts of many recordings.

Even without optimum settings, Sonic Foundry will usually not destroy the audio quality. In some cases you might readily enough hear a difference between its results and the best manual declicking, but you often won't know there should be any difference if you just listen to the Sonic Foundry result by itself.

Other people have recommended other plug ins and separate declicking programs. Some are rather expensive for a hobby (but still significantly less expensive than Cedar). I've not tried many of them. The Sonic Foundry plug in, for the first time, comes with Soundforge, that alternate audio editor. I don't know if it comes as the plug in, which could also be used by other DX compatible programs, or is somehow integrated only for use with Soundforge, but the Soundforge package is now less expensive than the plug in was by itself. As cliveb pointed out, there are also some good programs that work only on 16 bit files.

cliveb did not mention his own program, WaveRepair, that does provide ways for you to choose a replacement sample from elsewhere as one of its fix-it options. Clickfix's Force Fix operates by substituting samples from somewhere prior to the click for whatever you have selected for repair. Here you have no choice about what it uses, however. Sometimes the repair is seamless, sometimes it is quite unacceptable. You can use copy-paste to do the same thing in CoolEdit/Audition but you have to take care of more of the tricky considerations without program help.

As I mentioned in the first post above, the Syntrillium forum archive contain threads about the optimum use of CoolEdit's declicking. I'm certain there is some discussion of detection values, among many other aspects.

Also in there is a post I labeled ‘extreme declicking' which is a method I worked out for approaching declicking "perfection" that is sometimes less effortful than straight manual declicking (and will get more of the less audible clicks than most people will bother with when manual declicking). It provides a way to do automatic declicking but eliminate the removal of musical transients, and anything else not a real click. It is still quite labor intensive, so I've only used it a few times myself. There is another post, or three, in the same thread, about someone else's development of a way to get the best of several sets of declicking parameters on one track/album through use of multi-track.

Andy H-ha, cliveb and everyone else, I really do appreciate your "hanging with me" on this confusing topic. I have been doing a bunch of reading in the old Cool Edit archives and actually have already read the entire thread "extreme declicking" as well as the thread from the "multitrack" guy, who if I remember correctly, was from Down Under. I also now realize that there just isn't one simple method that will always work great with all vinyl recordings. I've also been doing quite a bit of experimenting with both the "Auto" and regular "Click/Pop Eliminator" in Audition. For now, I think I am going to try to stick with the simpler "Auto" version that only has two variables to change. Even though I do not really understand what the "Complexity" slider does, by keeping both parameters close to the preset "Light Reduction" I feel I probably won't mangle my music at least. My current goal is not visual waveform perfection, but simply to eliminate the majority of clearly audible clicks.

Regarding the more full featured "Click/Pop Eliminator" in Audition 2.0, I still think it's a shame that Adobe hasn't provided more explanation of how the process works and what the many parameters actually control. Trial and error can help, but randomly changing multiple variables without much direction is pointless. For example:

1. The graph that displays Detect and Reject Threshold values vs. amplitude has no units on the vertical axis. Are the Detect and Reject settings also amplitude values in dB?
2. All four of the preset curves have Detect values less than Reject. However, if you run the "Auto Find All Levels" button you generate curves that are just the opposite, Reject always less than Detect. Some of the curves are pretty strange looking as well. Does this even make sense?
3. The documentation states that when you raise the Reject value, it should reject more of the initially identified "clicks" that might be true music. In some of my testing, I had more rejections with a value of 40 than with a value of 50. What gives here?

I've decided that it is pointless to obsess over vinyl restoration. I'd like my music to sound better than no correction at all, but I really don't want to spend several hours on each LP. With that said, I'll ask just a few more questions:

1. Do people currently still like the Younglove method for crackle removal the best? If I use one of the scripts that are in the archives, does it really become a one button process?
2. Does anyone have an exact script to work with Audition 2.0? What I've found so far was written for CEP or perhaps AA 1/1.5.
3. How often do people use the very clever (but head spinning) "multitrack" method descibed by the bloke from Down Under?
4. Is ClickFix worth $50 as an addition to Audition 2.0? Audition was expensive enough and contains two separate options for click removal, even if they are rather poorly documented. Besides, I'm skeptical that ClickFix could more accurately find clicks by using "statistics" as stated in the FAQ. Doesn't the shape of the waveform have to scrutinized carefully?

Thanks again. I'm beginning to hate vinyl restoration and I have 723 albums still to go!

Why? I'm happy to convert all my LP to digital because the process almost fully automated. With a simple script in Audition 1.5 with ClickFix, it take me 5 minutes to process each track (Athlon64 3500+).

I'm considering go for faster CPU, also happy that I can hear the vinyl LP sound without wearing out the LP, it's much better than the CD, even I have a quite expensive "EAD UltraDisc 2000" CD Player.

My equipment for your reference, I know I'm far from perfect but that's all I can afford:

http://johnson.tmfc.net/vinyl

johnsonlam, that looks like pretty nice equipment. So why aren't you just using the declicking tools that are included with Audition 1.5?

Off topic: I was planning on peak normalizing to around -0.1 dB, like cliveb. But I just read a statement in the help files of Audition 2.0. It said if you're creating CD audio, don't normalize above 96% (-0.355 db) because some CD players make errors with music normalized to 100%. What do people think about this advice? Did this perhaps only apply to really old CD players and should have been deleted in the latest manual?

I find the ‘Younglove decrackling’ very useful for a couple or three different problems, on of those being some final declicking. It gets used on most albums I do. I have never done it via a script but my understanding is that Audition 2 scripts are different from those is earlier versions of the program.

Clickfix is very much faster than Audition’s declicking. I doubt that it identifies clicks with any greater accuracy. I haven’t read its literature for a few years, so I don’t recall what it says about detection methods, but I know for sure its correction method is different than Audition’s. It comes with some presets and ‘automated’ modes that take most of the guesswork and experimentation out of its use, further reducing necessary time spent.

I’ve seen more than a few posts from people who say they use it for all their declicking. I would guess that stems from two major factors. (1) it speed (2) it lack of complexity (although it does have eight parameters you could set yourself -- if you really though that would do any good). I don’t use it very extensively because I often don’t like the corrections it makes.

For some kinds of music, mainly that without low level passages, this fault might not matter, it might not be audible. Clickfix (and to a large extent CoolEdit too) leaves a significant portion of the low frequency part of the click intact. On much of the music I do, I can hear those left over stubs, and the effect is rather irritating, thus I don’t do that kind of declicking.

I’ve read that the 0dB problem is not found on any decent current CD players. I’ve normalized to 97% on the last 500 albums and have never noticed any problem. I would not put anything past the really cheap portable CD players, however.

Since I work almost exclusively in Spectral View, I haven’t bothered to try the multi-solution/multi-track approach. Has Audition 2 added a Spectral View to multi-track? It wasn’t available earlier.

By "manual pre-pass" I mean going through the WAV file by hand, seeking out the big clicks. I'm pretty obsessive about it, so I do actually listen through the whole file, usually in sections of 1 minute. I listen to a section, marking the clicks as it plays. Then I go back to each click and fix it by hand: switching to spectral view helps locate the exact position of the click. When that 1 minute section is done, I move on to the next section.

Methods of repair vary depending on the nature of the click/glitch. The most common methods I use are: Bezier and linear interpolations; spectral replacement (where the preceding and following sections are spectrally mixed then pasted over the glitch); block replacement (where a similar but undamged section is copied from nearby in the file); copying over the other channel (if it's undamaged and similar); smoothing (where each sample is replaced by the average of itself with the preceding and following sample); repainting the waveform using the mouse as a "pen"; adjusting the level downwards (typically by about 6dB or so); and in extreme cases actually deleting the samples when all else has failed (provided that doesn't make things worse).

The biggest logistical problem with doing it this way (if you plan to use an automatic declicker afterwards) is to decide whether a particular click is big enough that it needs to be manually fixed in order to prevent the auto-declicker from mangling it. That's a matter of judgement and comes with experience. As it happens, I tend not to use auto-declickers routinely, so I rarely need to make that particular call. LP restoration is a hobby, and I actually quite enjoy the task of manually repairing glitches.

If I do feel the need to use an auto-declicker, it's usually Wave Corrector, primarily because after its detection pass you can manually review its proposed fix for each click and reject those which are unacceptable. I'll also sometimes crank up CoolEdit 2000's cleanup plug-in, or (very occasionally) Groove Mechanic or Jeffrey Klein's ClickFix.

Somewhere else you asked whether ClickFix is worth the $50 cost. That's difficult to say, because I rarely use auto-declickers. On reflection I'd say it wasn't worth it for me. But if you want to get set up with auto-declicking, then it's one of the better packages out there, and for sure runs a heck of a lot faster than the native Click/Pop Eliminator. As the old saying goes, "you can't have too many tools" (although I must admit that my hard disk is overflowing with all sorts of audio restoration packages I've bought over the years, and most of them are now gathering dust).

Andy said that I hadn't mentioned my own program (Wave Repair). That's primarily because you said you'll be working at 32 bit, so it's not an option for you. But it might be a useful experiment to make a 16 bit copy of something and play around with Wave Repair. Its basic raison d'etre is to help simplify the tasks in manually restoring LPs. Everything you can do with Wave Repair, you can do with a general purpose audio editor, but those tasks which are common to fixing vinyl glitches are sort of "packaged" in Wave Repair so they are easier to execute. In playing with it, you might get some ideas about how to go about fixing glitches using Audition that auto-declickers can't handle.

One final comment. As you gain more experience, you'll find that your transfers get faster and better. After a while you'll go back to your first transfer and think to yourself, "how come I spent so much time on this and achieved such mediocre results?". In other words, be prepared to go back and re-do the earlier albums after you've got past the initial steep learning curve.

Can I share an opinion of mine that no one else seems to share?

I think the Cool Edit / Audition declicking is nearly useless. (Manual fix single click is good though)

Maybe if I spent more time with it, I could get it to work well. However, there are other programs which "just work". See Clive's page for suggestions.

Cheers,
David.

I declicked some number of albums with CoolEdit, possibly 50, that still sound good enough to me, but I wouldn’t want to have to depend on it anymore. It requires significantly more effort to avoid causing damage than is necessary to just find and treat clicks, although that depends quite a bit on the music. Some recordings are much easier than others.

I would assume that the clicks burried within music are the most difficult to remove accurately with an automatic declicker. As long as I can't hear these clicks, I'll be happy. When experimenting with the various settings in Audition 2.0, I have thus far been looking at quieter sections, often times between tracks. I've carefully looked here while in spectral frequency view. Sometimes there are extremely thin spikes that I really can't even hear, but I trust are clicks. I've had fairly accurate results eliminating these with the "Auto Click/Pop Eliminator" set on "light reduction". I gradually tweaked the two variables until the majority of these spikes were removed. The only thing that still concerns me is when it also removes spikes within the music that I can't even hear. I'm not experienced enough to know if these are real clicks or not. If the before and after music still sounds really good, I'm trying not to worry about these excised, inaudible spikes. (Though it is difficult! )


No, there is still no spectral frequency view while in multitrack view.


cliveb, you really should write a book detailing all the specifics you understand about WAV file processing. I will be your first pre-order!


You're going to be stranded with your computer and WAV files on a deserted island. You only have enough hard disc space for one declicker. What's your favorite? (And no, you can't just lay on the beach getting your tan.)

I've tried that, it still have clicks left.

I'm not very satisfy about the result and search the Internet for reference, I found ClickFix, it seems having a more positive review. I downloaded and give it a try, it's really good and price still acceptable.



I didn't really understand that statement. But I guess the Normalize it said is related to "peak", some of the "equalized" signal will be too loud after normalize, maybe the advise mean "don't push maximize the signal"

Just download a ClickFix and try it, don't pay until you satisfy with the result.

Audition built-in noise reduction is good, but click pop elimination not impressive (I mean Audition 1.5 or CoolEdit).

Indeed so, and dealing with this problem is what the Younglove technique aims to do. By isolating the low level noise & ticks, the auto declicker has a much easier job to do.


It is psychologically very difficult to accept what appear to be unnecessary changes to the waveform, but if they are inaudible, then try not to stress about it. (A corollary to this is that, when attempting to manually fix a glitch, you often need to extend the repair much further than the apparent visual disturbance to the waveform, even though the gut feeling is to edit as small a section as possible).


LOL. But if you're interested in some notes I wrote about the subject, see this webpage.


If it has to be an automated declicker, then I'd say Wave Corrector, or perhaps Sound Forge with Noise Reduction 2.0 plugin. If I'm allowed a manual declicker, then I am disqualified from commenting on the grounds that I have a vested interest.

I haven't kept up with what's best, so I'm still using the Sonic Foundry NR-2 plug-in declick. The actual declick part, not the vinyl restoration "do it all together" part. Medium noise, click shape 5-7, size 2-3 (lower to avoid artefacts), and threshold as needed.

It can introduce it's own artefacts (_all_ auto declickers can - even CEDAR) but real time preview makes it easier to tune the settings, and the worst artefact you'll get is a softening of the sound - especially transients. The worst artefacts of Cool Edit are quite horrendous in comparison.

I must stress that I have a very different approach to other people. I have tried the painstaking perfect / manual restoration approach in the past, but it's not for me. I am a perfectionist in many things, but I also like listening to music. Applying my perfectionist tendencies to audio restoration would mean that I never had time to listen to the music, and would have listened to it so many times while restoring it that I would never want to listen to it again when I'd finished!

There are great free demos available of most software though - that makes it easy to find something that suits you. I would strongly advise playing with some other pieces of software before spending hours tweaking Audition.

Cheers,
David.

That's the best piece of advice you've had in this entire thread.

Well, this is disappointing advice, though greatly appreciated. When I finally made the decision to "splurge" and buy Audition 2.0, I thought I was possibly buying the best. Now having used a small portion of it, I still think it is really pretty powerful. I think it was probably still a very good choice to help me record and process my albums. It definitely could use better documentation, but I'm gradually learning many things by trial and error. But I think I will experiment with other software, as long as there is a free trial period.

Now for some exciting news! Last night I created my first CD from my first vinyl record! (I know, you're thinking big deal.) I did the entire process in Audition 2.0 and I think it actually turned out great! I didn't do all the processing steps that many people suggest. I simply wasn't sure that all the steps were necessary. Here's what I ended up doing:

1. Clean album with Nitty Gritty vacuum.
2. Engage subsonic filter on preamp.
3. Record to laptop using an Echo AudioFire 2. This is a firewire device that I set for 44.1 KHz/24 bit.
3. Create a single WAV file in Audition 2.0 stored as 32 bit float file.
4. Run the auto click/pop eliminator process on entire WAV with custom settings close to the "light reduction" preset.
5. Listen very carefully to the entire WAV with Grado headphones while marking potential missed clicks.
6. Zoom in on the potential clicks in Spectral Frequency mode. Select these clicks with the "marquee selection tool".
7. Run the "favorites" process "repair transient".
8. Carefullly listen to before and after.
9. Normalize WAV to 97%.
10. Trim some silence from beginning, between sides and at end of WAV.
11. Place track markers 20 CD frames before each song begins. (That's 266 ms. I read suggestions of 200-300 ms.)
12. Convert WAV to 44.1 KHz/16 bit with dithering enabled. I followed AndyH-ha's suggestions of 0.5 bit depth, triangular pdf and C1 noise shaping.
13. Insert WAV into CD view and burn, baby burn.

I know. Where was the crackle removal, noise reduction and rumble filtering? I may rumble filter if I actually hear a problem. I think my subsonic filter should eliminate very low frequencies such as record warp. I didn't really hear any crackle (plus I'm afraid of Younglove.) I'm certain I'm going to have to master this method for other albums. Noise reduction? I don't think I heard noise, but I'm also not entirely certain that I'd recognize it. Perhaps I should have done an automatic noise reduction on all the between track intervals. But, auto click/pop eliminator seemed to quiet these sections quite well. They were pretty quiet to begin with however. Despite this minimalist approach, the CD sounds great! Track selections work like a charm, including one segue where I had to place my track marker 1/2 frame before music started.

I do still have a question about the proper way to manually eliminate clicks. When in spectral frequency mode, I could sometimes identify a very faint click that did not completely go the the top or perhaps the bottom of the display. Other times, the click might clearly extend from the top, but then merge into the bottom, brighter colors and be completely indistiguishable from the other low frequencies. It may be that these clicks were already partially processed when I declicked the entire WAV file. Anyway, I delt with these by precisely selecting only the visible abnormality with the marquee tool. I didn't include low frequencies or the very top frequencies when I couldn't discern the spike. I also did only the left or right channel if it wasn't present in both. I then ran the repair transient favorite.

Would it be better to avoid the marquee tool and instead just use the time selection tool so that all frequencies are always selected? Also, should I actually be selecting a few "milimeters" of normal waveform on both sides of the click? What do you routinely do?

cliveb, I know you've stated that you often times will only do a manual click removal, but that you enjoy your hobby. I just can't imagine doing the manual removal first. It would take me forever. I'm hoping that by auto declicking first, my manual touchup might be quicker. But if this approach is less accurate, I suppose I could change for album number two.

It is quite common, maybe we should say ‘the norm' for auto declicking to only partially remove a click. Fortunately that is often adequate but is one of the reasons I don't often use this particular program's auto declicking. All the programs I've used have this characteristic, but the audibility of the remainder varies.

Turn on Synchronize Cursor Across Windows (under Options). Open the unprocessed copy of your recording. Now, when you have a suspicious or possibly partially processed click, switch windows to the unprocessed file to see what it was before auto declicking. Sometimes it is very useful to select a section containing whatever was mangled, copy, then paste it into your in-process file, good as new.

There is no single way to use the manual declick, which happens to be one of the most useful tools in the entire program from the ‘restoration' viewpoint; its uses are legion, for much more than just clicks. My version does not have the marquee tool, but while that is probably a very useful option, it sounds much too time consuming to be worth while in most cases.

I almost allways correct each channel in separate manual operations (not at different time periods).

The three major variables for manual declicking are
exactly what you select
the correction parameters
how many times you ‘correct'

selection
This has to be learned by experience through trial and error. It is often the case, especially on very large clicks, that selecting only one sample in the click is the route to the best result. This necessitates switching to waveform view in order to see the exact form of the click and to choose the particular sample.

Fortunately it generally it isn't necessary to go to that extreme to get undetectable removal. One can stay in Spectral View and just perform a select on the click as displayed. Generally, it is desirable to be zoomed in to around 100 milliseconds full screen, sometimes closer, for greatest control, but of course that is more effort, so you won't always want to go that far in.

For the majority of clicks, place the cursor in the center of the brightest part (often dead center in the Spectral View, but not always). However, there are places, especially in vocals, where your repair works better is you select along the outer edges of the click.

Press left arrow (or shift_right arrow) once to select a slice through the display of the click. Apply the manual fix (Audition 2 calls it something other than "Fill Single Click Now" no?).

This small selection works best on the majority of clicks, but sometimes you need to select more, sometimes a region around the click, possibly 100 milliseconds or more, especially when the click has an extended low frequency component. There are often clues to that in Spectral View, but you may need to switch to waveform view to see what is what. Trial and error until you easily recognize most of what you need to know.

The three Correction parameters are
(1) FFT Size
(2) Pop Oversamples
(3) Run Size

I get away with three sets but I'm sure there are problems that would be easier with some different values. I've assigned three favorites to function keys so I can just press the keys to execute, without ever going into the dialogue.

Most frequently used, by far: (1) 128 (2) 20 (3) 25. I used to mostly use 256 for parm (1), it is more through, but 128 gives me finer control. I haven't bothered with 256 in quite a while. (With FFT Size 256, auto correction tends to leave less low frequency stub).

Next most common is (1) 512 (2) 20 (3) 25. This is mostly needed when a larger selection is required but can be a follow up to the first fix, without changing the selection.

Rather infrequently used is (1) 8 (2) 2 (3) 1, useful only for special, very limited results.

times to apply
Pressing my function key once is often adequate but many clicks require two or three ‘fixes,' sometimes more, usually without changing the selection. Very large clicks may require selecting a number of different critical samples in succession.

Sometimes, especially with larger clicks, or those with nasty low frequency thumps, I get the best results through whittling down the mess with the 128 size correction, narrowly applied, multiple times, followed by a larger selection and the 512 size fix (occasionally applied two or three times). After this last correction it can be desirable to Amplify before you loose the selection. Try amplifying by 2 or 3 dB.

In a beginning CoolEdit 2000 tutorial, before every using the tools, I read that Spectral View settings
Windowing Function: Blackman Harris
Resolution: 256 bands
Window Width: 40%
were best for identifying clicks.
Use the Logarithmic Energy plot. The default 120 dB is usually best.
Reverse Color Spectrum Direction should be checked.

While you can play with the colors of Waveform View until your heart is content, changing Spectral View colors will only lead to harder to see clicks.

Other values are useful for some other kinds of problems -- the most common variation is increasing the Resolution -- but for declicking, the above is probably best. "Probably" because I can't say how big a factor it is that these settings are what I became used to before I ever experimented with anything else.

Decrackling is not needed if there is no crackle. However, the Younglove technique is actually good for more than decrackling. I often use it as the last step of declicking (preceeded by one to four auto declicking runs, using different parameter values in each run, two runs being most common). Another use of Younglove is to reduce the effects of some distortin on badly abused disks.

Noise Reduction
I've written in two or three threads in this forum about the proper use of Audition's Noise Reduction (also that of the Sonic Foundry NR 2 plug in). Results from awful to very good depend upon the settings. For some reason, the program documentation is very wrong on this.

Many albums need no noise reduction but more often can be improved (to my tastes, anyway) by NR applied only to the lower frequencies, somewhere around 500Hz and below. Also, a little light NR to the full frequency spectrum sometimes makes the music a little better, perhaps clearer, even though the noise isn't really audible per se.

pre declicking
If clicks are large enough to be visible (in waveform view) when the entire album is on screen, they have a low probability of being processed very well by auto declicking. Some albums have no such problems, some have many. Being able to paste the original back in, and thus start over, as described above, is a work-around to the mess that auto declicking sometimes makes of large clicks.

I never listen to the entire thing from that viewpoint before auto declicking, but I always do at least a little investigation in order to decide what auto declicking I will try. If I run across very loud clicks that don't show up as mentioned above, especially ones that are very bright as the base in Spectral View, I know auto declicking isn't likely to do an optimum job.

When two, or ten, clicks are virtually on top of each other, auto declicking is pretty useless and manual preprocessing is best. Again, trial and error will teach you what you need to do when.

No need to stress about this: Audition probably IS the "best" general all-round audio editing program. (Although I've never used Audition per se, only CoolEdit Pro and CoolEdit 2000. I would sincerely hope that any changes Adobe have made will have been for the better). Some people might prefer Sound Forge, but I think that's more a matter of personal reaction to the style of their interfaces. Last time I played with Sound Forge, I got the feeling that CoolEdit (hence Audition) had a more comprehensive set of facilities. Two things in Sound Forge's favour IMHO are: (i) the Noise Reduction 2.0 plug-in performs a little better, and much faster, than CoolEdit's Click/Pop Eliminator; (ii) Sound Forge has a "pen" tool for redrawing waveforms, which is more useful than CoolEdit's "drag sample value up/down" feature (or has Audition added a "pen"?).


Well done. Probably the most important step in getting a good transfer (assuming the turntable is set up correctly).


Apart from his (her?) scary name, there's nothing to be afraid of. The Younglove technique is pretty benign, certainly much less likely to create artifacts than an auto-declicker applied to the actual music signal.


If I ever use noise reduction, I tend to limit it to quiet sections, ie. inter-track gaps and quiet parts of the music. In louder sections the music signal usually masks the noise. One downside to applying it only to the quiet sections is that you have to be careful about where it is introduced, otherwise there can be a noticable change in the noise background when it cuts in. I usually apply noise reduction in a level-dependent manner (the higher the level, the lower the amount of noise reduction). This means the reduction in background noise gets introduced gradually and there's no sudden change.


I'm not familiar with the "marquee tool" - it must be something new in Audition. If I use CoolEdit's "Fill Single Click" feature, I typically select the full extent of the visible glitch in spectral view. Since spectral view "smears out" the visible glitch, selecting it this way usually guarantees that it encompasses the entire glitch.


Don't use my personal approach as a role model - I'm probably quite weird in this respect (ie. actually enjoying the process itself)! But if you want the best possible results, then my opinion is that a full manual pass will be required, whether or not you also use an auto-declicker.

To echo what Clive said, I'm still using Cool Edit Pro 1.2a and it is essential for me. It's just the automated declicker that I don't use.

btw, since you've mentioned it, I don't use the CEP denoise very much either because IMO NR-2 is superior - but the CEP denoiser is still good. (Audition 2.0 may have improved in this respect.)

I agree with the others on denoising LPs - sparingly, usually selectively, if at all. Less is more.

One big problem I went through is that I'd do audio restoration listening through headphones, and then listen to the finished CDs via speakers. Bad move. The results sound fine, but headphones are so revealing that I'd spend a lot of time and effort removing problems which would never be audible through speakers anyway. If you want to be able to enjoy the result via headphones in a quiet environment, then you'll have to put a little more effort and care in. You can do a lot of damage by doing "too much" and this is more obvious via headphones too!

Congratulations on your first CD. If you enjoy listening to it, then you've done something right!

Cheers,
David.

Yes, I've came to the same conclusion. I usually apply NR only to the lower frequencies, up to about 250 Hz. But it depends greatly from case to case.

Great suggestion! I would have never thought about cutting and pasting. I used this yesterday on my second album. I discovered a short segment where the auto declicker had actually decreased some fast snare drums. I cut and pasted from the original WAV and presto, repaired file!


Did you really mean 100 ms? That's only 1/10 of a second for the entire width of the monitor. My spectral frequency display looks extremely stretched and blurred when zoomed in that much.


Many times you just place a very narrow selection within the actual width of the click? I'm surprised that there would be any click removal outside the selection. I thought all processes only took place within actual selected waveform.


The marquee tool allows you to draw rectangles around any point of the waveform while in spectral frequency view. There is also a free form "lasso tool", but I think it is more difficult to control. Occasionally my clicks have not covered the full height of the spectral frequency display. I've seen some of AndyH-ha's lower frequency "stubs" and other clicks that start in the high frequencies, but don't quite reach the very lowest frequencies. For these clicks I have been just "boxing" the visible glitch, instead of using the time selection tool. It seems to work fine, but I wasn't sure if I should really be using only the time selection tool for the "fill single click" option.


Yesterday I downloaded ClickFix to give it a try. I've only attempted a few repairs. I selected an in-between tracks segment that was full of discrete clicks in spectral frequency view. None of the presets, including "33 1/3 Aggressive" or "Auto Aggressive" removed them. They only faded partially. In contrast, Audition's "light reduction" setting removed everyone of them. I was able to get ClickFix to also remove them by adjusting the sliders. But I'm surprised that the aggressive presets were of no real help.

I also downloaded a demo of the Sound Forge Noise Reduction 2 plugin. I haven't tried it yet, but I believe it can be used within Audition. However, so far, I haven't seen it available for separate purchase except on Sony's site at full retail. I could buy the entire Sonic Forge 9 program elsewhere for less than just the plugin from Sony. Though, I believe Sonic Forge would still cost around $250, uugghh!


You know, it really does sound very good! It was great yesterday listening to it in the car. It is a fairly obscure album from 1975 that I'm sure is simply not available for purchase on CD.

I set the zoom to approximately 500–1000ms. And zoom in additionally for somewhat trickier parts.
Spectral settings affect the display. For manual declicking I set the FFT size to 256 and use Blackmann windowing function with width set to 40%. It gives (at least to my eyes ) a nice sharp spectrogram suitable for declicking.

Mid-70's, eh? That's my era. You'd be surprised how many obscure albums from that period actually *are* available on CD. What album is it?

OK, I'll accept your challenge. Not only is it not available on CD, I'll bet you've never even heard of the artist. He was one of my more rare rock discoveries that was never played on the radio. I believe I own all of his albums, totaling three. Are you ready? (......................long drum roll.....................)

David Werner!

There's a slim chance you heard of his first album called "Whizz Kid". The one I recorded is his second and actually better called "Imagination Quota". The third album, also great, is simply called "David Werner".

OK now, be honest! Did I stump you?

Guilty as charged: never heard of him. I found a brief review on the net of his third album, which describes him as a bit like Bowie, and far too talented to be unknown. So not having heard his stuff is probably my loss.

100 milliseconds is quite common during my manual declicking phase, although I never think about or look at those numbers except when I want to write about the process. There are times when I need to be twice as close, and even more, for optimum control. Did you notice my suggesting for Spectral View settings? The particular settings you use have a major impact on what you see.

If you had the energy to read my last post very closely, you should find mention of the fact that selecting only one sample (which requires about 1 millisecond full screen zoom) is sometimes the route to the best fix, meaning the way to achieve the least adverse impact. While this is quite dependent on the click and the context; I usually know from experience when to do what, but undo is a good thing to be familiarly with.

I don't often go to the extreme of selecting only one sample, as the differences between that and a larger selection are often minor, and there are many instances requiring multiple samples, but I do employ the selection technique I described with cursor placement and arrow key, giving the smallest possible selection for the current zoom level, for 90+% of clicks.

In waveform view, zoom in close enough to be able to select a single sample at the peak of a sharp click. Now zoom out enough to see the context. Apply the ‘fix' and observe how much of the waveform changes. The amount of change is very dependent on the correction parameters; usually the FFT Size 128 settings makes the most modification.

I think the Sonic Foundry plugin may only be available via download from their website. The lower price for the entire package is what I meant in my earlier post. The plug in definitely works from within CoolEdit/Audition but I don't know if the Sound Forge package has something different that cannot be used elsewhere.

Not only is the Sonic Foundry declicking easier than CoolEdit's/Audition's but I also use its NR more often.

cliveb has mention before what he wrote in his last post
I usually apply noise reduction in a level-dependent manner (the higher the level, the lower the amount of noise reduction)

I understand a series of stepped applications, each with a different setting, but if you know a way to apply varying amounts of NR in a single pass, I would appreciate a description. This is a bit like the difference between using calculus to compute the area under a curve vs employing geometry to approximate it with a series of polygon calculations.

AndyH-ha, I tried these settings but didn't like them. I wanted to make sure that we were looking at similar displays. Here is a screen capture of the start of a record with clicks using the above settings. Under "spectral controls" there were a few other settings left at default: Saturation 75, Gamma 2 and Transparency 40. (Reverse and Logarithmic are not defaults in Audition 2.0.)



Here is a screen capture of the display I've been using. The settings are as follows:

Windowing Function: Blackman Harris
Resolution: 512 bands
Window Width: 100%
Range: 132 dB
Linear plot
Not reversed
Saturation: 75
Gamma: 1.4
Transparency: 40



Am I using the wrong display? I know that for different monitors or lighting situations, these settings are personal preference. But the reverse and logarithmic settings look confusing to me. If you have the time, I'd love to see a screen shot of your preferred spectral frequency display for identifying clicks.

I did read that, but I couldn't believe what I was reading. I have zoomed in to the "sample level" for fun, but I don't think I could recognize a click this magnified.


I tried this, but must not understand how to do it. Do you have to hold your mouse button while hitting the arrow key to affect a drag? If I just clicked and then pressed the left arrow, it didn't seem to create a selection, just the initial click point. I tried "fill single click" but got the message that I needed to select something.

If you're using a typical audio editor such as CoolEdit, then I only know of the "stepped" approach. After I decided that a level-dependent application is appropriate, I added it as an option in Wave Repair. (As I've mentioned on a number of previous occasions, the intention of my software is to simplify the typical editing procedures used when resoring LP recordings. When I find something that works well, I make it as easy as possible in Wave Repair if I can).

Good deal about the WaveRepair NR. Noise reduction is one of several things I've often wanted to be able to apply on a curve.

Spectral View
The difference in appearance between 256 and 512 bands is not large at lower zoom, but the larger number of bands stretches individual transient events over a longer duration, contributing to your blurry picture at 100 milliseconds. Just a little experimentation also show me that the screen refresh isn't quite as good at 512 but
(1) you may well have a newer and more powerful computer/video system that handles scrolling better
(2) I've read some comments about Audition 2's Spectral View being generated in a new way, relative to prior programs versions, that lets the program render it much faster.

100% Windowing blends together some close clicks that show as separate, but very near, disruptions when using 40%. Less low frequency detail is visible. This also makes the screen content less distinct at close zoom.

Your settings may work well for you but, as I already wrote, more precise selection can often lead to less destructive changes. It all depends on the context and the click itself. You can view the differences more easily by zooming in to 100 milliseconds near some clicks and changing settings.

A range of 132dB rather than 120dB only means that fainter, i.e. smaller, things will show up more clearly. Turn it up to 200dB for a real gasp as what a mess of noise you have. Showing smaller, and thus less important, clicks and noise is sometimes very useful but, in general, I view it as just a distraction that I don't want to be remained about.

Your preferred screen seem to have the same basic color scheme and general appearance as mine. Is there a lot of green and dark blue when you select something? To really see if there is any difference worth noting, except as I described above about Resolution and Window Width, we would need the same bit of the same recording on screen. Changes to Saturation, Gamma, and Transparency are not available in my version, so I can't say if I would prefer any different values.

I don't at all understand your first screen shot. If that the result of making the plot logarithmic? Must be due to a significant program change. If I change mine to a linear plot, a great deal of detail vanishes, making the view mainly black and white. Possibly I could get used to it for grosser declicking, since only the larger clicks even show up, but that isn't an appealing prospect.

Zooming In
In Spectral View it is hard to identify many things when zoomed in enough to select individual samples. When I need that kind of selection, I locate the item of interest at a lower zoom, place the cursor, and zoom in by clicking on the zoom button (or by using cursor keys, depending on my mood). I change to waveform view either before or after the zoom. Some clicks are still easy to identify in waveform view at that high zoom, but others change too slowly for enough of them to actually appear on screen at one time -- you kind of have to go on faith because you placed your cursor before you zoomed in so close.

100 milliseconds is truly useful for good control, especially for significant clicks in the middle of some vocals (varies a lot by singer) and some instruments (brass, in particular). You may need to go in to 10 milliseconds in order to get only the click and not greatly effect the transient of the rapidly repeating waveform. You might note that I also wrote that the 100 millisecond screen wasn't generally necessary. 300 milliseconds full screen is often close enough to be very useful. By the time the screen width gets out to 900 milliseconds, it is harder to easily place the cursor exactly where I want it without fiddling (unless I need to select a region rather than a single event), and it is not possible to make really small selections.

Cursor Key Selection
To select with arrow keys, click the mouse button to set the cursor position. Left arrow selects towards the left, right arrow moves the cursor right. Shift left arrow moves the cursor left and shift right arrow selects towards the right. You can enlarge the selection by continuing to move the cursor with the arrow keys. It is quite difficult to get the same degree of control by dragging the mouse.

If you can't select this way, chalk up another change for the program. Does the cursor move? Surely they wouldn't eliminate basic cursor key control. Perhaps they made it more logical by always requiring the shift key be down (or some other key?) in order to select, and the arrow key alone now always just moves the cursor. The variation in functioning between going right and going left in my version always seemed rather peculiar.

X-Ray,

Re: Andy's settings - I think you've put the log scale into effect on the wrong thing. He is plotting log intensity. You're plotting log frequency. You don't want to do that - it'll make the clicks harder to spot.

For click spotting, I would use 128 bands, whatever dB range makes the clicks stand out (e.g. 100dB), and don't reverse the colour spectrum. I get green plots with bright green clicks.

For general audio analysis I use something completely different. 512-1024 bands (to see the frequencies / notes clearly) with 140dB range (to see the background noise clearly).


FWIW, after the automatic declicking I've described, I can't see many clicks. This isn't the case with other declickers. What you see has some correlation with what you hear. YMMV.

Cheers,
David.

I've done quite a bit of audio cleanup of vinyl (300+ lps). Even so, this discussion has reached technical levels I didn't think possible :-).

For what it's worth here is my experience:

I used CoolEdit 2000 with the audio cleanup plugin. I played with the different presets as well as the auto detect capability. I found that the "medium level" was a, well, happy medium for me.

I created some scripts within CE2K and was able to automatic it where I could feed it a big batch of wav files and it would chew on them overnight.

In my experience, I found 2 types of music (actually instruments) that would regularly result in noticibly distorted audio:
- saxaphone, for instance Spyro Gyra or David Sanborn.
- synthesizer, for instance Peter Baumann's "Romance '76" or Edgar Froese's "Solo 1974-1979". Some Vangelis caused problems too. A more well known example would be Floyd's Dark Side of the Moon (especially the low frequency passages in On the Run). I never converted DSOTM, but it contains synth passages like those that I've seen cause problems.

I think the sonic nature and frequency range of these instruments would trip up the algorithm and cause much of the content to be filtered thinking it was clicks. The results would sound like audio playing through a bad electrical connection (e.g. a tape deck connected to a receiver using patch cables with corroded connectors). When I first experienced it, my first step was to check my electrical connected between my stereo and PC. The resulting audio would lack punch. It was so bad, I would revert to the original unprocessed version and burn that to CD.

Anyway, that's my experience. I had a lot of fun with it, especially developing my batch procedure. Though I didn't go into the level of detail you guys are getting into.

Cheers.

The main weirdness of that first screen shot results from "reversing" the color spectrum. Audition 2.0 defaults to yellow>red>purple, with yellow being the highest amplitude frequencies and purple the lowest. When I click the "reverse" button it shows all those very low amplitude frequencies as bright yellow.


You're absolutely right. You can tell in the first screen shot that by clicking the "logarithmic" button, it changed the vertical scale to logarithmic frequency. My second screen shot shows "linear" frequency. Perhaps Audition 2.0 has changed. I cannot find anywhere to make the color spectrum logarithmic. I assume that is what you're suggesting. Also, my default color scheme for spectral frequency display goes from yellow to purple. I cannot really appreciate any green color at all, though it may be there. The only time I clearly see green is in waveform display which is set on green.


OK, I've finally figured it out. Thanks for the method. Left click places the cursor. Right or Left arrow keys will move the cursor a tiny amount. Shift-Arrow selects a tiny amount, either right or left.

Thanks for this warning. I actually do own a lot of jazz/new age on vinyl such as Spryro Gyra, David Sanborn, Dotsero, Images, The Rippingtons, Bob James, Earl Klugh, etc. I also own lots of music with synthesizers. I'll be very careful when applying any sort of clean up on these albums.

I'll confirm this warning. Auto declickers tend to trash brass and reed instruments (so sax perhaps is a double whammy, being both brass AND reed), and "raspy" synth patches. If you take a look at the shape of the waveform transient for these kind of instruments, it's very similar to the "instant-up, gradual fade" shape of a typical LP click.

When you select a portion of your Spectral View screen, the color stays the same?

Since I have no place of my own for images, I used the third post here
http://www.hydrogenaudio.org/forums/index....mp;#entry511037

Audition 2 offers transparent selection in the Spectral View. Although you can check 'Invert' in the Spectral controls and have the display as in your picture.

My default is for the spectral colors to stay unchanged, with a faint, transparent selection window over them. As eevan stated, "invert selection" makes it look like your screen (though I didn't understand that setting until now. Thanks eevan.)

some details are easier to see when inverted.